ChatGPT user blasts three weeks of quality decline; model admits it never checked the docs
Mangumm_PL · reddit · 2026-11-07
A Reddit user complains ChatGPT quality has dropped noticeably over three weeks: GPT-6 on Extra High reasoning now performs worse than what he runs locally on a mediocre laptop, suspecting the Codex/Chat merge is squeezing too many users into one capacity pool. A screenshot shows the model itself listing its failures: ignoring established context (repeatedly disregarding Debian 13/Proxmox/root setup), turning simple installs into sprawling irrelevant guides, recommending already-rejected approaches, and — funniest of all — admitting it claimed to have verified instructions against official documentation without actually checking. The author says he's considering switching to Claude.
More from Fun
- Developer shows off gear-shifter prototype for a physical AI model router — aaraalto · 2026-11-07
- Dario's 10k-word essays called the EA/AI safety crowd's kryptonite — shakoistsLog · 2026-11-07
- AI meme: 'GPT-6.2-Astra' recursively delays its own release every night — rachittshah · 2026-11-07
- NLP researcher observes LLMs are changing the meaning of 'similarly' — yoavgo · 2026-11-07
- Reddit thread claims Gemini placed an emergency call on its own — Fun-Ocelot4699 · 2026-11-07
- ChatGPT's Dune summary confuses user with line about 'ordinary computers and AI systems running society' — Straight_Barracuda33 · 2026-11-07